JPS5893099A - Voice recognition circuit - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice recognition circuit in a voice response system, and more particularly to a voice recognition circuit that can efficiently correct erroneous voice recognition results.

When a voice response system, for example, a dial telephone, is used to inquire about a balance at a specific bank, the bank's computer uses a voice recognition circuit to respond to the caller (hereinafter referred to as the speaker).
It recognizes the voice uttered by the computer, converts the received voice into numbers, etc., and inputs it into the computer. And the computer
Then, the bank account balance corresponding to the account number is returned to the speaker via the telephone line, and the speaker confirms it with a synthesized voice from the computer. In this way, in addition to the case where the response is made remotely using a rotary telephone, the response is made by synthesized voice or characters displayed on the monitor, even when the response is made by directly operating the terminal.

In this system, the speech recognition circuit may misrecognize the voice uttered by the speaker, but if the utterance confirmation voice from the computer is i@tb, the speaker confirms it and You must input your voice again based on the instructions from the user.

In the conventional Jinnotane voice recognition system, it is necessary to perform operations until an accurate recognition result is obtained, but that is not possible. There are 6 to 9 tests that take a long time. Furthermore, if the recognition rate is not good and erroneous recognition occurs frequently, the speaker must input the same voice over and over again, which causes pain to the speaker.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech recognition circuit that eliminates the drawbacks of the prior art described above and can efficiently correct erroneous recognition even when it occurs.

The feature of the present invention is to prepare in advance the statistics of words that are likely to be misrecognized in the voice of the speaker, and to output that one word in response to the misrecognition when it is misrecognized. The person who achieved this is also G.

Hereinafter, one embodiment of the present invention will be described in detail with reference to the attached drawings.Figure 0 is a schematic configuration diagram of a voice recognition system, which is connected to a dial telephone via a voice matching line and a central control unit (CPU). In Figure 0, which is an example of exchanging information with a telephone, the audio input terminal 1 corresponds to the output from the telephone's transmitter. 2 is a speech recognition unit, and a speech analyzer 3. A verification circuit 4 that performs verification between the audio signal from the word table 6, which will be described later, and the audio signal obtained through the audio analyzer 3. When the voice is verified, the "Hi" voice is input and the command to open the gate at 9 o'clock is sent to the voice Kl@f! When the sense is 1L, the gate is closed when the "Yes, J" voice is input. It consists of a gate opening/closing control circuit 5. 7 is the gate circuit, and the output terminal 8 is connected to the CPU.

The word table 6 stores words that are statistically likely to be mistaken, and when the word that is likely to be mistaken is input and determined to be a mistake, the word table 6 reads out the corresponding word. For example, as shown in the figure, statistically
gt #) The easy words are 2L, S81, and 12 as well.
The words that are easily mistaken for '1' are memorized as t1#, %81, and when the single i1 '1' is negated in the speech recognition unit 2, the words 2' and 8' are used as the next recognition result. This is for automatic output.

Further, numeral 9 is a speech synthesizer for returning the registered words read out from the word table 6 to the speaker. lO
is a speaker, which corresponds to a telephone handset.

Although the figure only shows the voice recognition circuit system, voice recognition can also be performed by voice and commands from the CPU. is omitted to simplify the explanation.

In this circuit, multiple voices inputted through the input terminal 1 are analyzed by the voice analyzer 3 of the recognition section 2, and then verified by the verification circuit 4. This result becomes a response voice by the voice synthesizer 9 and is returned to the speaker through the speaker 10. At this time, if the recognition result is correct, input the predetermined input word "high" to operate the gate opening/closing control circuit 5 of the recognition unit 2, opening the gate circuit 7 (the contact is closed in the figure). , and output the recognition results of the input speech through the output terminal 8.

If the aforementioned response voice returned to the speaker differs from the input voice, the predetermined input word "Iie" is returned.
is input, the gate opening/closing control circuit S of the recognition unit 2 is operated, the gate circuit 7 is closed (the contact is opened in the figure), and it is determined that the previously inputted audio word was misrecognized. , the word ``2'' is read out from the word table 6, which stores in advance statistically erroneous words ``r = J, ``r hachi'' corresponding to the recognized word ``ichi'', and the entanglement circuit 4. The signal is output from the speaker 10 via the voice synthesizer 9. If this is correct, by inputting the ``hi'' voice to the input terminal l, the word ``two'' is outputted through the gate circuit in the same manner as described above. Also, if the word ``two'' is also incorrect, please input ``yes'' next. It is output from the speaker 10 via the voice synthesis circuit 9.

The timing for outputting the next recognition result may be immediately after the negative input (no) of the first recognition result, or when the previous negative word is recognized when inputting again.

As mentioned above, by using the statistical properties of words, words that are similar to the word are stored in a word table in advance,
When the recognition result is negative, we read out nine approximate words from the word table, confirm them by voice, operate the gate circuit, and send them to the CPU. Since it performs word recognition, highly efficient word recognition can be achieved.

In addition, in the above-mentioned embodiment, numerical values were given as examples of words that are easy to atb and words that are similar to them.

Speech recognition is not limited to this, but can be quickly and easily achieved using a similar circuit system for alphabets, katakana, and even simple words.

As is clear from the above explanation, according to the present invention, the statistical properties of words are used to provide an approximate word table and read it out when a mistake is made. This has the effect of achieving highly efficient speech recognition.

[Brief explanation of the drawing]

The attached figure is a speech recognition circuit diagram showing one embodiment of the present invention. 2...Speech recognition unit, 3...Speech analyzer, 4...Verification circuit, S...Gate opening/closing control circuit, 6...Word table, 7...Gate □ gate circuit, 9...・Speech synthesizer,
10...Speaker.

Claims (1)

[Claims] A voice recognition circuit in a voice response system includes a word storage means for storing in advance a plurality of words that are statistically likely to be misrecognized to correspond to words in input speech that are likely to be misrecognized, and a speech synthesis circuit that recognizes the input speech. voice recognition means that reads another word from the word storage means when the recognition result obtained through the axillary voice recognition means is misrecognized; and gate means that outputs the voice information only when the iI recognition is achieved in the axillary voice recognition means. A speech recognition circuit comprising: